NEW YORK — Achronix Semiconductor Corp., eight years into its startup phase, is hanging its hat on Intel’s 22-nm FinFET process technology for its survival and envisioned victory as it prepares to go public in 2014.
Achronix’s 2010 decision to shift its foundry business from Taiwan Semiconductor Manufacturing Co. to Intel’s fledgling foundry operation was a bold move that surprised many in the semiconductor industry, for two reasons.
First, when Achronix and Intel announced the deal, Intel had zero experience as a foundry; Achronix was literally Intel’s first customer. Today, however, Intel is said to have several undisclosed chip customers—including some big ones—for its foundry business, and it has publicly announced two other customers: PLD maker Tabula and programmable network processor provider Netronome.
Second, Achronix is likely paying a whole lot more to buy wafers from Intel than from TSMC. How can the startup afford it, and when does it expect its gambit to pay off?
“We think Intel’s 22-nm process will put us 2-½ years ahead of our competitors,” Achronix founder and chairman John Lofton Holt said in an interview with EE Times this week. Indeed, Holt said, the startup is so confident in the partnership that “we are now investing in Intel’s 14-nm process technology.”
Intel is “so far ahead now [in the industry on the process technology] and they know it,” the Achronix chairman said. He called Intel’s decision to enter the foundry business “a brilliant move,” adding, “They’ve got the fabs. Why wouldn’t they monetize them?”
Holt is betting the partnership will be Achronix’s ticket to the big leagues in high-density, low-power FPGAs. With its newest Speedster22i HD and HP families produced on Intel’s 22-nm process, Achronix hopes to elevate its game for head-to-head competition with Altera and Xilinx.
Achronix is not sampling its 22-nm chips to customers yet, but Holt said that it has gone through three tapeouts with Intel and that it has chips on hand. By the fourth quarter, he said, Achronix expects production to be ramped up, and it intends to announce several customers by name for the first time.
It took Achronix and Intel only 12 weeks from their first meeting to strike their foundry agreement, Holt recalled. “We couldn’t announce the deal until November 2010, because we had to be in Intel PR machine’s queue,” he said.
Asked whether his investors had objected to the idea of contracting with a fab that had never served as a foundry, Holt answered to the contrary, saying, “Fortunately, they also saw it as a potential grand slam.”
Selling Achronix’s engineers on the idea took a bit more finessing. “Engineers are fundamentally conservative,” Holt said. “They were worried at first, but they are on board now.”
Holt cited three reasons for his confidence that Achronix’s decision to work with Intel on 22 nm would outweigh the risks.
First, the partnership brings Achronix access to hardened intellectual property. Intel was willing to port its IP, including PLL, I/O and memory blocks, to the 22-nm process Achronix is using. As a result, Achronix has integrated hardened-IP support for Ethernet MAC, 100G Interlaken, PCI Express and DDR3 controllers around the edges of its high-density, high-performance FPGAs, giving it a leg up on power consumption, according to Holt.
With its earlier products, Holt said, Achronix “underestimated how important power is to our customers,” even when its FPGAs were to be used in big-iron telecom/network equipment. “Low power consumption is our big focus.”
Second, Achronix benefits from Intel’s halo effect, which allays potential customers’ concerns about everything from the startup’s staying power to the quality of its chips.
Holt said the effect has been evident in Achronix’s recent experiences with the global supply management groups at tier-one telecom and networking companies. GSM groups assess potential suppliers for their adherence to quality, accounting and best-practices guidelines, Holt said, and “the end result of the GSM analysis is a ‘thumbs up’ or ‘thumbs down’—usually a thumbs down—that gates a vendor's ability to sell to the company.” While GSM groups are understandably startup-averse, Holt said, “in our case, most GSM groups have been very bullish about working with us because they see Intel as our ‘big brother.’ ”
Intel’s support for everything from packaging to qualification and supply chain management has been helpful, he added.
The third factor, Holt said, is cost. Without offering specifics, he acknowledged that the company is paying Intel more than it paid TSMC. But Achronix expects to leverage Intel’s 22-nm process to offer its customers large, high-end, low-power FPGA chips on the volume market at half the of competitors’ equivalent products. That will more than compensate the startup for its higher outlays for Intel-manufactured wafers, he said.
While Achronix has had to justify its decision to become Intel’s first foundry customer, Intel’s decision to work with Achronix didn’t take a lot of thought, according to Holt. Long after the deal was signed, Holt said, the vice president of Intel’s foundry business told Holt that Intel had considered Achronix a low-risk customer. Intel had contracted to produce big chips that were difficult to make, but it would be doing so in low volumes for a startup client. If Intel had struggled in its first foundry effort and Achronix had gotten burned, there would have been less heat reflected back on Intel than there would have been if its first customer had been, say, a Broadcom or Qualcomm.
There has been some speculation that Achronix might be looking to an Intel acquisition offer as its exit, but Holt insisted getting acquired was “not in our plan, not for the foreseeable future.” The endgame, he said, is an IPO and “we are only two years away from that.”
Nonetheless, Holt acknowledged that using Intel’s process technology opens the door to integration of one’s products with Intel’s products, and suggested that Intel customer Tabula might play that angle.
Combining Intel’s process technology with Tabula’s Spacetime 3D architecture—“assuming Spacetime works,” said Holt—would let Tabula serve a consumer to mid price-range market with its products. If successful, Tabula could become an Intel acquisition target.
But of course, that’s just a speculation at this point.
Why would they bother with Achronix. They would have to relearn RTL development as Achronix forces a new methodology on the designer in order to get performance. No FSMs allowed.
Besides, Intel can just fab their own test chips, why bother with FPGA emulation?
Couple of reasons not to get too excited:
1) Intel's using Achronix as a guinea pig for their process technology. There's no monetary investment from Intel in Achronix. So if Achronix fails, don't look for Intel to bail them out.
2) Achronix is a tiny company. They're charging mucho $$$ to get into their early access program (i.e customers pay to be guinea pigs for Achronix). GA is not until next year -- but based on their track record -- more like 2014.
3) You have to pay extra (a lot extra) to use their hard IP. Basically all the SERDES, Memory and PCI interfaces are hard IP. So just to use the chip normally you have to pony up big $$$. Not to mention you could by a nice server for the price of the chips themselves.
4) Their programming model doesn't allow for feedback loops if you want the high performance. Guess what a Finite State Machine is?
This company has gotten a lot of attention from the Intel deal, but it's a dud as a viable FPGA company.
"word on the street" was that Intel's ground rules are pretty conservative since they basically only make microprocessors. And FinFETs can't help that much.
And if the process technology is *truly* that superior, you have eliminated any possibility of a second source, ever.
So I'm not sure it's so clear cut to use Intel as a foundry. But I'm just an engineer, we don't usually get to make decisions, just have to execute on them.
Intel is using Achronix to tune their new processes - what better than a relatively homogeneous circuit architecture, no pressure on volumes, and someone to pay you to do that?? I am sure Achronix know that and they are going along with it for the reasons mentioned in the article. If their gamble pays off, happy days, if it does not, well they've live in the limelight for a while...
From Acronix's point of view the pros would definitely outweigh the cons. They get access to Intel's technology and support. They also have a potential exit strategy if Intel ever decides to acquire them. For Intel, they get to increase their fab utilization and increase their diversification. The con for Acronix would be that Intel could increase their wafer cycle time if the fab becomes full. Mostly I would say that it is a win-win for all except TSMC.
Actually a low-volume case like Achronix could screw up the smooth wafer flow in the affected fab, with dedicated wafers and masks. Intel's fabs run smoothly in the context of high volumes, i.e., wafers containing many identical dies, not mixtures of limited number dies.
Join our online Radio Show on Friday 11th July starting at 2:00pm Eastern, when EETimes editor of all things fun and interesting, Max Maxfield, and embedded systems expert, Jack Ganssle, will debate as to just what is, and is not, and embedded system.